Personal Heart Monitoring System Using Smart Phones To Detect Life Threatening Arrhythmias

Transcription

1 Personal Heart Monitoring System Using Smart Phones To Detect Life Threatening Arrhythmias Peter Leijdekkers, Valerie Gay Faculty of IT, University of Technology Sydney UTS FIT, PO box 123, Broadway 2007 NSW Australia [peterl, Abstract This paper discusses a personalized heart monitoring system using smart phones and wireless (bio) sensors. We combine ubiquitous computing with mobile health technology to monitor the wellbeing of high risk cardiac patients. The smart phone analyses in real-time the ECG data and determines whether the person needs external help. We focus on two life threatening arrhythmias: Ventricular Fibrillation (VF) and Ventricular Tachycardia (VT). The smart phone can automatically alert the ambulance and pre assigned caregivers when a VF/VT arrhythmia is detected. The system can be personalized to the needs and requirements of the patient. It can be used to give advice (e.g. exercise more) or to reassure the patient when the bio-sensors and environmental data are within predefined ranges. 1. Introduction Cardiovascular disease is regarded as Australia s greatest health problem. In 2003 it killed 48,835 people in Australia [5]. In the developed world cardiovascular disease is the leading cause of mortality. With an ageing population it will only increase over the next 10 to 15 years since cardiovascular deceases occur predominantly in the age group of 55 and over [4]. Risk factors for developing cardiovascular diseases can be medical but also lifestyle related [4]. Major risk factors are diabetes mellitus, high blood pressure and high blood cholesterol. Obesity also contributes to developing cardiovascular deceases. Obesity is becoming a major problem in developed countries and the World Health Organization (WHO) estimates that globally over 1 billion adults are overweight and at least 300 million of them are obese [24]. This problem is related to physical inactivity which is another factor contributing to developing cardiovascular diseases. In this paper we propose a portable monitoring system that monitors the heart and notifies the person or external party in case of abnormalities. Our monitoring system is meant for patients that have a known cardiovascular disease and need to be monitored around the clock. Traditional heart monitoring solutions exist for many years such as the Holter device which records the patient s ECG for 24 to 48 hours and is then analyzed afterwards by the cardiologist. The patient can wear the device and go home and resume his/her normal activities. The main drawback of these solutions is when a major incident occurs during the monitoring phase. This is recorded but no immediate action is taken to help the user. Other solutions have been introduced that address this problem and J.Rodriguez et al [14] have classified these solutions in two groups. The first group uses smart phones (or PDAs) equipped with bio-sensors that record the heart signals and transmit them to a health care centre or hospital for analysis. Some solutions can store the signals locally as well. Examples include Alive technology [3], Vitaphone [2], Ventracor pocketview [6] or Welch Allyn Micropaq [7]. Most are capable of recording, viewing and storing ECGs directly on the smart phone. Some solutions transmit the stored ECG to the health care center using wireless technologies (e.g. GPRS). The second group aims at building platforms for realtime remote health monitoring. Examples are Mobihealth [10], TeleMediCare [17], OSIRIS-SE [19] and PhMon[18]. These solutions use (wearable) wireless sensors to monitor patient s vital signs (e.g. ECG, oximeter, blood pressure). The European project Myheart [16} develops such a platform and focuses on heart patients. Myheart aims at designing intelligent biomedical clothes for monitoring, diagnosing and treatment. The platforms developed by this second group collect the bio data and send it to a care-centre or a hospital for processing and analysis.

2 None of these solutions process the ECG data locally on the smart phone, and the ECG signals need to be continuously transferred to a health center if the patient needs to be monitored 24/7. This can be costly when using GPRS transmission. To deal with this issue several research projects consider processing the ECG data on a local device such as the Amon, Epimedic and Molec project. AMON [13] is a wrist-worn medical monitoring and alert system targeting high-risk cardiac and respiratory patients. The system includes continuous collection and evaluation of several vital signs, smart medical emergency detection, and is connected to a medical centre. For heart monitoring, they are technically limited by the fact the device is worn on the wrist and therefore the ECG signal is very noisy and not suitable to diagnose cardiac abnormalities. Epi-medics project [15] defines an intelligent ECG monitor which can record, analyze the ECG signals and can generate alarms. It can also be personalized but it is not a device meant to monitor the patient 24/7. The patient connects to the 12 lead monitor periodically as directed by the heart specialist or when he/she doesn t feel well. MOLEC [14] provides a solution that analyses the ECG locally on a PDA. It generates alarms to the hospital in case of high risk arrhythmias. Our objective is to investigate and develop an application whereby a heart patient is monitored using various types of sensors (ECG, accelerometer, blood pressure monitor). The sensor information is collected and transferred wirelessly to a smart phone. Our solution is similar to AMON, MOLEC and Epi-medics and analyses the ECG on the local device. One distinction of our solution compared to the others is that we can personalize the monitoring and we have mechanisms in place to locate the user in case of an emergency whether the patient is indoors or outdoors. We detect life threatening arrhythmias and give the patient general information about their health when they are not in a dangerous situation. We can also store extra information for further use by health providers. This paper presents our work to date on a personalized heart monitoring system. Section 2 summarizes the main requirements for our solution. Section 3 gives an overview of our solution and describes the main components of our prototype. Section 4 concludes on the status of our project and discusses open issues. 2. Requirements One of the challenges was to determine what information is essential and what is secondary for high risk cardiac patients. The first step was to identify what to collect, what to process locally, what to store for further use and what needs to be personalized What needs to be collected? Using a smart phone and bio-sensors we have the potential to collect a lot of data. Numerous bio-sensors exist that provide health related information such as ECG, blood pressure, sugar level or physical activity. Other important data include context information such as location, outside temperature which can have an impact on the wellbeing of the patient. We can also obtain data from the user such as the current weight, whether he/she is smoking or not. We can maintain a Patient Medical Profile and store contact details in case of an emergency. For high risk cardiac patients, the ECG signal is the obvious data that needs to be collected continuously and should be given priority over all the other sensor data. It is also important to store the ECG signal for further analysis by the cardiologist. Detecting falls is another important indication that the patient is in danger. At the time of the heart attack many patients collapse and an accelerometer sensor can be used to detect falls. Finally, in case of an emergency we automatically want to call an ambulance and direct it to the location of the patient since many patients black out or are unable to speak. A GPS sensor can be used outdoors to accurately obtain the location of a patient. However, GPS does not work indoors and we need to complement it with other location sensors such as the GSM Cell ID or WiFi access point locations. With GSM Cell IDs and WiFi access points we are able to provide a rough indication of the location of the patient as described in [11] What needs to be personalized? From literature and discussions with cardiologists we identified that personalization is an important aspect since the threshold levels and what needs to be monitored vary for every patient. For example, some cardiac patients need to monitor their sugar level as well, whereas other patients need to monitor their weight and blood pressure. Also threshold levels for raising an alarm differ depending on the patient s age and condition. The frequency of the monitoring and what needs to be stored for further analysis varies per patient and is determined by the cardiologist.

3 The level of physical activity recommended for a heart patient depends on his/her condition and health history. National Heart Foundation of Australia [9] says that physical exercise improves the live expectancy of heart patients and they set guidelines to help heart specialists setting a personalized level of activity for their heart patients. Using an accelerometer/pedometer and other contextual information, we can evaluate the level of activity of the heart patient. We assess this level of activity against the heart specialist s personalized guideline and either congratulate the patient for reaching his/her goal or to encourage them to exercise a bit more. Since our target group will be mainly elderly people, the interaction with the monitoring application needs to be personalized and adapted to the user s health condition (e.g. bad eye sight, hearing problems). For example we need voice interaction in case the patient has bad eyesight or vibration and flashing lights for hearing impaired patients Quality of the data collected The monitoring system is only useful if we know the quality of the data we receive from the various biosensors and the quality of the diagnosis based on that data. Knowing the quality we can put mechanisms in place to compensate for the lack of accuracy of certain sensors or diagnosis. For example, before raising an alarm, we might crosscheck whether the patient is in danger by asking a few questions or perform an extra reading on a particular sensor. We plan to develop a simple voice-activated cognitive function test that can be initiated by the smart phone after the monitoring application detected a potential fall. It is also important to provide accurate but yet nonoverwhelming information to the patients since we do not want to cause extra anxiety which would make the situation worse for a heart patient. For this reason we do not show an ECG diagram to a patient since we learned from discussions with cardiologists that this is a major source of anxiety for cardiac patients. 3. Prototype Figure 1 shows a simplified view of the heart monitoring system. The heart patient has one or more wireless sensors (e.g. ECG, accelerometer, pedometer, Oxygen) attached to his/her body. The sensor information is collected and transmitted to the smart phone. We use off-the-shelf technology enabling us to incorporate the best sensors as they appear on the market. The sensors we use are Bluetooth enabled or integrated into the smart phone (e.g. GPS, WiFi). The smart phone processes the sensor data and monitors the patient s wellbeing and in case of an emergency it automatically calls an ambulance to the location of the patient. It can also warn caregivers or family members when the patient is in difficulty. The smart phone stores information that can be transferred to the patient personal health record via the internet. Figure 1 Personalized Heart patient monitoring We developed the personalized heart monitoring application on Microsoft s Windows Mobile Pocket PC platform. We selected this platform due to easy access to lower level APIs which are needed for the sensor manager modules. Also the tight integration with the operating system allows easier access to other applications running on the mobile device such as the calendar application, WiFi and obtaining the GSM Cell ID. We used the.net Compact Framework extended with OpenNETCF [12] modules to build the application. The data is stored in an SQL CE Server which is a compact database for mobile devices. During the first phase of our project, we focused on the sensor managers and the smart heart monitor Sensor managers Each (bio) sensor manager gathers and processes the data from a specific sensor to determine a diagnosis. The results and accuracy of the findings are forwarded to the smart heart monitor for further interpretation. Depending on the patient, the specialist can configure one or more bio-sensors to be used to monitor the patient. The specialist configuration is password protected and is only accessible by a medical specialist. For each sensor, we can set the polling frequency and we have implemented security features that will automatically reconnect in case of connection loss and low battery detection so that the user can recharge the battery. As discussed in the previous section ECG, accelerometer and GPS are the core sensors for our application. We use a Bluetooth GPS sensor (Emtac)

4 and an integrated ECG/Accelerometer bio-sensor from Alive Technologies [3]. We selected this bio-sensor since it has been demonstrated in [1] that the Alive ECG sensor provides reasonably good signals for detecting normal or abnormal arrhythmias. Additionally, the Alive accelerometer has been used successfully to detect falls in a study involving stroke patients at the Prince Charles Hospital (Australia) [1]. ECG sensor manager ECG signals can be a source of errors which makes it hard to interpret the correct arrhythmia. In our prototype we work with a two lead ECG sensor. Noise, interference and non-rest conditions of the patient can contaminate the signal. This implies that we focus on extreme ECG signals. In the first stage of the prototype we focus on two life threatening arrhythmias: Ventricular Fibrillation (VF) and Ventricular Tachycardia (VT). VF is a lethal arrhythmia characterized by rapid, chaotic movements of the heart muscle that causes the heart to stop functioning and leads quickly to cardiac arrest. VT is an abnormal heart beat usually to a rate of beats per minute. VT may result in fainting, low blood pressure, shock, or even sudden death. ECG and check it for a VT/VF rhythm. We implemented a VT/VF detection algorithm as detailed in [21]. If the algorithm detects either a VT or VF signal the emergency procedure is started. Figure 2 (right) shows the ECG configuration form where thresholds can be set for a patient. Heart rate thresholds are set to either notify the specialist or initiate the emergency procedure. GPS sensor manager As stated before GPS is only useful when the patient is outdoors and in clear sight of GPS satellites. We foresee that many heart patients will spend most of their time indoors and in order to automatically determine the location we used WiFi and GSM as a means to determine the location when indoors. Since GSM Cell ID and WiFi access points are not automatically related to a location the user has to relate a particular location with the WiFi/GSM Cell ID data. Figure 2: ECG sensor specialist configuration To detect these arrhythmias we implemented and adapted a beat detection and classifier algorithm initially developed by EP Limited [23]. This algorithm is able to detect and classify a heartbeat as Normal, PVC 1 or Unknown. We determine the heartbeat rate which will be checked against the thresholds set by the cardiologist for the patient. If the rate is too low or too high the application will inform the specialist/user or start the emergency procedure (see section 3.2). If we deal with a PVC or unknown beat we will record the 1 Premature Ventricular Complexes (PVCs) are extra heartbeats [20]. PVCs interrupt the normal heart rhythm and cause an irregular beat. PVCs are often harmless, but when they occur very often or repetitively, they can lead to more serious rhythm disturbances. Figure 3: Location configuration Figure 3 shows several screenshots how a user can automatically spot WiFi access points and GSM Cell IDs and assign these to a particular address. If GPS is available the longitude/latitude coordinates are also assigned to the address. When an alarm is raised the location manager will automatically start sensing the environment for WiFi access points and GSM Cell IDs. If a match is found, the related address will be used as the current location of a patient. For this scenario to work, the user needs to sense and input the addresses where he/she is normally staying Smart Heart Monitor The Smart Heart Monitor component receives the results from the bio-sensors and determines whether an alarm should be raised. As stated before the results of the bio-sensors can be inaccurate due to noise and inaccurate readings. The smart heart monitor will

5 therefore access the results of the bio-sensors and in case the quality is under a certain threshold level, it needs to crosscheck whether the patient is in danger to avoid raising false alarms. In the current implementation we collect extra data before raising an alarm from the particular bio-sensor. In case we still measure a life threatening situation we show an alarm form and play a recorded message notifying the user (Figure 4, left). The user can disable the alarm if it is a false alarm. Otherwise a first aid message is played continuously on the smart phone instructing (potential) bystanders what to do in case the patient is unable to speak. Simultaneously, an emergency call is placed automatically by the application. In our prototype we chose to implement three warning levels (Figure 4, right): Red means that immediate intervention is required. The Yellow warning level indicates that certain threshold levels are reached and is used to warn the user or specialist. Finally, Green indicates that everything is normal. We deliberately kept the interface simple and only show the heart rate and the status of the sensor(s). Additionally audio is used to warn the user of an abnormality. For the ECG we only show the heart rate and not an ECG diagram. The reason is that patients do get anxious when showing detailed ECG information and it might worsen the situation. The smart heart monitor module has access to information related to the patient such as emergency contacts. Figure 4: Emergency form (left), Yellow warning screen (right) Our solution is meant to monitor the patient continuously and battery life of the used devices is an issue. The ECG sensor battery lasts for approx 60 hours. The smart phone s battery only lasts for approx 8 hours when continuously connected and processing data. This can be an issue if the wearer is not close to the charger. However, studies show that many heart patients are sedentary and can therefore charge the smart phone while being monitored. 4. Conclusion This paper described a 24/7 personalized heart monitoring application using smart phones and wireless wearable bio-sensors. We developed a working prototype focusing on ECG sensors and developed a heart beat classifier complemented with a Ventricular Fibrillation and Ventricular Tachycardia detection algorithm. The algorithms have been tested on ECG files stored in the MIT/BIH Arrhythmia Database [22]. This is a commonly used database to test and study different types of arrhythmias. The initial results are very promising and detailed test results will be published in a forthcoming paper. With these algorithms we are able to detect life threatening arrhythmias and automatically call for external help. By processing the ECG data locally on the smart phone we can supervise a patient without being continuously connected to a health-centre. This reduces the workload of medical staff, communication costs and motivates the patient s self-care. We can also store data for further analysis by the specialist or for inclusion in the patient electronic health record. The ECG sensor is small (match box size) and can be easily worn without being noticed by other people. The smart phone is relatively large compared to an ordinary mobile phone but due to the wireless communication no cables are needed to connect the sensors to the smart phone. The target audience for our application is patients that have had a heart attack, or are at high risk. We learned from discussions with cardiologists that these people are worried that a heart attack will happen again and are very motivated to wear a device that can monitor an reassure them. Intrusiveness seems not to be an issue for these highly motivated patients. We are currently focusing on the personalization aspect in cooperation with cardiac specialists and we are refining which information should be displayed depending on the type of patient. Another research topic is to accurately determine whether a patient has fallen or not. Furthermore we plan to test our application on cardiac patients in cooperation with a Sydney hospital to determine the feasibility of our prototype. This will provide us with valuable feedback to improve the application. We believe that our system is a non intrusive heart monitoring application which will promote patient s autonomy and by providing personalized advice we hope that it will give the patients more confidence and improve their quality of life.

6 5. References [1] Boyle J., Wark T. and Karunanithi, M., Wireless Personal Monitoring of Patient Movement and Vital Signs, e-health Research Centre, Australia, IEE CIMED 2005 Proceedings. ISBN: , [2] Vitaphone, [last accessed 7 th April 2006]. [3] Alive Technologies, [last accessed 7 th April 2006]. [4] American Heart Association. Heart Disease and Stroke Statistics 2005 Update. Dallas, Texas.: American Heart Association; [5] Australian Institute of Health and Welfare Secondary prevention and rehabilitation after coronary events or stroke: a review of monitoring issues. AIHW Cat. No. CVD 25. Canberra: Australian Institute of Health and Welfare. [last accessed 7 th April 2006]. [6] Ventracor pocketview, [last accessed 7 th April 2006]. [7] Welch Allyn Micropaq, ess/micropaq.asp [last accessed 7 th April 2006]. [8] Briffa T., Maiorana A., Sheerin N.J., Stubbs A.G., Oldenburg B.F., Sammel N.L.and Allan R.M., Position Statement : Physical activity for people with cardiovascular disease: recommendations of the National Heart Foundation of Australia, Medical Journal of Australia Volume 184 Number 2, January [9] Physical Activity Recommendations for People with Cardiovascular Disease 2 [last accessed 7 th April 2006]. [10] Jones, V. et al. MobiHealth: Mobile Health Services based on Body Area Networks. In: M-Health Emerging Mobile Health Systems. Springer-Verlag, Berlin, pp , 2006, ISBN [11] Leijdekkers, P. and Gay, V., Personalized Service and Network Adaptation for Smart Devices, IEEE APCC Asia Pacific Conference on Communications Perth, Australia, October [12] OpenNETCF.org, The Premier.NET Compact Framework Shared Source Site [last accessed 7 th April 2006]. [13] Anliker, U.et al. AMON: a wearable multiparameter medical monitoring and alert system, IEEE Transactions on Information Technology in Biomedicine, Volume 8, Issue 4, Dec Page(s): [14] Rodriguez, J., Goni, A and Illarramendi, A., Real-time classification of ECGs on a PDA, Information Technology in Biomedicine, IEEE Transactions on, Volume 9, Issue 1, March 2005 Page(s): [15] Rubel P. et Al, Towards personal ehealth in cardiology. Results from the EPI-MEDICS telemedicine project, Journal of Electrocardiology 38 (2005) pp [16] Myheart [last accessed 7th April 2006]. [17] Telemedicare dicare/ [last accessed 7th April 2006]. [18] PhMon Personal Health Monitoring System with Microsystem Sensor Technology [last accessed 7th April 2006]. [19] OSIRIS-SE Runtime Environment for Data Stream Management in Healthcare [Last accessed 7th April 2006]. [20] Premature Ventricular Contractions, American Heart association, r=4695 [last accessed 7th April 2006]. [21] Ayesta U., Serrano L., Romero I.,Complexity Measure revisited: A new algorithm for classifying cardiac arrhythmias, 23rd Annual International Conference of the IEEE Engineering in Medicine and Biology Society (2001). [22] PhysioBank, ANSI/AAMI EC13 Test Waveforms, [last accessed 7th April 2006]. [23] EP Limited, Open Source Arrhythmia Detection Software, [last accessed 7th April 2006]. [24] World Health Organization (who), Obesity and Overweight, cts/obesity/en/ [last accessed 7th April 2006].